SPECjEnterprise2010 & Java Enterprise Edition (EE) PCM Model Generation DevOps Performance WG Meeting

Transcription

1 SPECjEnterprise2010 & Java Enterprise Edition (EE) PCM Model Generation DevOps Performance WG Meeting Andreas Brunnert Performance & Virtualization Group, Information Systems Division fortiss GmbH An-Institut Technische Universität München

2 Agenda SPECjEnterprise2010 Architecture Orders Domain Workload Benchmark Execution Procedure Pitfalls Automatic Performance Model Generation Data Collection Data Aggregation Model Generation Next Steps 2 DevOps Performance WG

3 Agenda SPECjEnterprise2010 Architecture Orders Domain Workload Benchmark Execution Procedure Pitfalls Automatic Performance Model Generation Data Collection Data Aggregation Model Generation Next Steps 3 DevOps Performance WG

4 SPECjEnterprise2010 Architecture Image Source: [2] Details about components, db tables: 4 DevOps Performance WG

5 SPECjEnterprise2010 Architecture Orders Domain (CRM): Java EE Web Application Manufacturing Domain Enterprise JavaBeans (EJBs) driven by the Orders Domain and a special Manufacturing Domain Workload to produce cars Supplier Domain (SCM) Web-Services used by the Manufacturing domain to order parts required during the manufacturing process no specific driver only indirectly Image Source: Brunnert, Andreas; Vögele, Christian and Krcmar, Helmut (2013): "Automatic Performance Model Generation for Java Enterprise Edition (EE) Applications." In: Computer Performance Engineering 10th European Workshop on Performance Engineering, EPEW 2013, Venice, Italy, S DevOps Performance WG

6 SPECjEnterprise2010 Orders Domain Workload Image Source: Brunnert, Andreas; Vögele, Christian and Krcmar, Helmut (2013): "Automatic Performance Model Generation for Java Enterprise Edition (EE) Applications." In: Computer Performance Engineering 10th European Workshop on Performance Engineering, EPEW 2013, Venice, Italy, S DevOps Performance WG

7 SPECjEnterprise2010 Benchmark Execution Procedure SPECjEnterprise uses the Faban Benchmarking Harness to specify the workload and to execute load on the system Most important input variable: Scale (Needs to be at least 50) Scale specifies: The number of users; Scale*10 Their behavior - in terms of how many entitities they access during the execution The number of items in the database 7 DevOps Performance WG

8 SPECjEnterprise2010 Benchmark Execution Procedure SPECjEnterprise2010 checks the SUT configuration (e.g. database content, caching behavior of JPA) before every benchmark execution!!! Therefore, one always needs to reload the database before executing a benchmark! Afterwards, the Benchmark can be executed however, one still needs to specify a Ramp-Up, steady state and Ramp-Down phase for the benchmark execution. This time depends on: Scale/Number of Users JVM settings such as Xms/Xmx The minimum ramp-up time can be calculated based on the delay that is specified between thread-starts (default: 100ms) We recommend ~10 minutes for ramp-up (even for up to 4300 users) and a thread start delay of 100ms. 8 DevOps Performance WG

9 SPECjEnterprise2010 Benchmark Execution Procedure 9 DevOps Performance WG

10 SPECjEnterprise2010 Pitfalls Not fully loaded databases: SPECjEnterprise2010 allows a small tolerance for the database content This means that one is able to re-run a benchmark if a previous one has been canceled after a few minutes (as not many items have been modified) However, this might change your results Different JPA implementations (Hibernate vs. EclipseLink) Check your persistence.xml carefully to see if caching is disabled in the correct technology specific - syntax Stateful JPA Some JPA Implementations (e.g. the one in JBoss ) are quite sensitive to schema changes, so two reboots might be required to get the atomicity tests running 10 DevOps Performance WG

11 SPECjEnterprise2010 Pitfalls Full garbage collections during the benchmark execution Either make sure that the memory is big enough to handle one benchmark run or let the benchmark run long enough to reduce the impact of single spikes in the results For Linux: set all language and time settings to en_us!!! Otherwise you might not be able to see the graph results in your benchmark harness Use systems with the ability to measure time intervals below 10 ms (e.g. CenOS, opensuse, Solaris, AIX, ) and not Windows or Ubuntu as you might get incorrect results See dissertation of Michael Kuperberg for more details Make sure that the hostname in the Driver configuration matches the one in hosts file of the driver OS 11 DevOps Performance WG

12 Agenda SPECjEnterprise2010 Architecture Orders Domain Workload Benchmark Execution Procedure Pitfalls Automatic Performance Model Generation Data Collection Data Aggregation Model Generation Next Steps 12 DevOps Performance WG

13 Automatic Performance Model Generation Java Enterprise Edition (Java EE) Typical Java EE Application Architecture (Example): Components according to the Java EE Specification: Servlets / Java Server Pages (JSP) handle requests/responses [2] Enterprise Java Beans (EJB) contain Business-Logic Database contains persistent data 13 Content Source: Brunnert, Andreas; Vögele, Christian and Krcmar, Helmut (2013): "Automatic Performance Model Generation for Java Enterprise Edition (EE) Applications." In: Computer Performance Engineering 10th European Workshop on Performance Engineering, EPEW 2013, Venice, Italy, S [2] Image Source: Neubig, Stefan (2014): "Evaluierung der Ressourcenabschätzung einer beispielhaften Unternehmensanwendung auf Basis eines automatisch generierten Performance-Modells." Bachelor's Thesis, Technische Universität München Performance fortiss GmbH DevOps Performance WG

14 Automatic Performance Model Generation Data Collection To represent these predefined component types the following data needs to be collected Servlets, JSPs, EJBs names Component operation names [2] Component relationships on the level of their public operations Resource demand of these component operations External systems such as databases 14 Content Source: Brunnert, Andreas; Vögele, Christian and Krcmar, Helmut (2013): "Automatic Performance Model Generation for Java Enterprise Edition (EE) Applications." In: Computer Performance Engineering 10th European Workshop on Performance Engineering, EPEW 2013, Venice, Italy, S [2] Image Source: Neubig, Stefan (2014): "Evaluierung der Ressourcenabschätzung einer beispielhaften Unternehmensanwendung auf Basis eines automatisch generierten Performance-Modells." Bachelor's Thesis, Technische Universität München Performance fortiss GmbH DevOps Performance WG

15 Automatic Performance Model Generation Data Collection To represent these predefined component types the following data needs to be collected Servlets, JSPs, EJBs names Component operation names [2] Component relationships on the level of their public operations Resource demand of these component operations JDBC Wrapper External systems such as databases 15 Content Source: Brunnert, Andreas; Vögele, Christian and Krcmar, Helmut (2013): "Automatic Performance Model Generation for Java Enterprise Edition (EE) Applications." In: Computer Performance Engineering 10th European Workshop on Performance Engineering, EPEW 2013, Venice, Italy, S [2] Image Source: Neubig, Stefan (2014): "Evaluierung der Ressourcenabschätzung einer beispielhaften Unternehmensanwendung auf Basis eines automatisch generierten Performance-Modells." Bachelor's Thesis, Technische Universität München Performance fortiss GmbH DevOps Performance WG

16 Automatic Performance Model Generation Data Collection To represent these predefined component types the following data needs to be collected Servlets, JSPs, EJBs names Component / Operation name Component operation names [2] Component relationships on the level of their public operations Resource demand of these component operations JDBC Wrapper External systems such as databases 16 Content Source: Brunnert, Andreas; Vögele, Christian and Krcmar, Helmut (2013): "Automatic Performance Model Generation for Java Enterprise Edition (EE) Applications." In: Computer Performance Engineering 10th European Workshop on Performance Engineering, EPEW 2013, Venice, Italy, S [2] Image Source: Neubig, Stefan (2014): "Evaluierung der Ressourcenabschätzung einer beispielhaften Unternehmensanwendung auf Basis eines automatisch generierten Performance-Modells." Bachelor's Thesis, Technische Universität München Performance fortiss GmbH DevOps Performance WG

17 Automatic Performance Model Generation Data Collection To represent these predefined component types the following data needs to be collected Servlets, JSPs, EJBs names Control flow (ID + Call Order + Stack Depth) Component operation names [2] Component relationships on the level of their public operations Resource demand of these component operations JDBC Wrapper External systems such as databases 17 Content Source: Brunnert, Andreas; Vögele, Christian and Krcmar, Helmut (2013): "Automatic Performance Model Generation for Java Enterprise Edition (EE) Applications." In: Computer Performance Engineering 10th European Workshop on Performance Engineering, EPEW 2013, Venice, Italy, S [2] Image Source: Neubig, Stefan (2014): "Evaluierung der Ressourcenabschätzung einer beispielhaften Unternehmensanwendung auf Basis eines automatisch generierten Performance-Modells." Bachelor's Thesis, Technische Universität München Performance fortiss GmbH DevOps Performance WG

18 Automatic Performance Model Generation Data Collection To represent these predefined component types the following data needs to be collected Resource demand: CPU end CPUstart Servlets, JSPs, EJBs names Heap end Heapstart Component operation names [2] Component relationships on the level of their public operations Resource demand of these component operations JDBC Wrapper External systems such as databases 18 Content Source: Brunnert, Andreas; Vögele, Christian and Krcmar, Helmut (2013): "Automatic Performance Model Generation for Java Enterprise Edition (EE) Applications." In: Computer Performance Engineering 10th European Workshop on Performance Engineering, EPEW 2013, Venice, Italy, S [2] Image Source: Neubig, Stefan (2014): "Evaluierung der Ressourcenabschätzung einer beispielhaften Unternehmensanwendung auf Basis eines automatisch generierten Performance-Modells." Bachelor's Thesis, Technische Universität München Performance fortiss GmbH DevOps Performance WG

19 Automatic Performance Model Generation Data Collection To represent these predefined component types the following data needs to be collected Servlets, JSPs, EJBs names Performance Curves which capture #of calls and response time within the scope of specific requests Component operation names [2] Component relationships on the level of their public operations Resource demand of these component operations JDBC Wrapper External systems such as databases 19 Content Source: Brunnert, Andreas; Vögele, Christian and Krcmar, Helmut (2013): "Automatic Performance Model Generation for Java Enterprise Edition (EE) Applications." In: Computer Performance Engineering 10th European Workshop on Performance Engineering, EPEW 2013, Venice, Italy, S [2] Image Source: Neubig, Stefan (2014): "Evaluierung der Ressourcenabschätzung einer beispielhaften Unternehmensanwendung auf Basis eines automatisch generierten Performance-Modells." Bachelor's Thesis, Technische Universität München Performance fortiss GmbH DevOps Performance WG

20 Automatic Performance Model Generation Data Aggregation The Data Collection and Model Generation Components can work with either work with InvocationBasedPersistence Services (Files) or AggregatedPersistenceServices (MBeans, DBs) Java EE Server Data Collection MBeans Model Generation Files Data Aggregation Analysis DB Image Source (adapted): Brunnert, Andreas; Vögele, Christian and Krcmar, Helmut (2013): "Automatic Performance Model Generation for Java Enterprise Edition (EE) Applications." In: Computer Performance Engineering 10th European Workshop on Performance Engineering, EPEW 2013, Venice, Italy, S DevOps Performance WG

21 Automatic Performance Model Generation Model Generation Java-based client Usage Model Based on Data: Stored in Runtime Mbeans Repository Model Stored in Files or an Analysis Database Default repository, system, allocation and System Model resource environment are generated automatically PCCurve Specification file is also generated Allocation Model automatically and is also associated with the system model and its corresponding repository Resource Environment model interfaces Image Source: Wischer, Kilian (2014): "Abbildung von Ressourcen-Profilen für Unternehmensanwendungen durch Performance-Modelle." Master's Thesis, Technische Universität München DevOps Performance WG

22 Automatic Performance Model Generation Model Generation Image Source (adapted): Brunnert, Andreas; Vögele, Christian and Krcmar, Helmut (2013): "Automatic Performance Model Generation for Java Enterprise Edition (EE) Applications." In: Computer Performance Engineering 10th European Workshop on Performance Engineering, EPEW 2013, Venice, Italy, S DevOps Performance WG

23 Agenda Introduction Motivation Automatic Performance Model Generation Data Collection Data Aggregation Model Generation Evaluation SPECjEnterprise2010 Deployment Model Generation Measurement and Simulation Results in Comparison Related Work Outlook 23 DevOps Performance WG

24 Outlook Usage Model Generation Master Thesis ongoing based on session traces Setting thread limits for each tier (Web, EJB, JDBC Conn. Pools) Representing Distributed systems Distributing the transaction ID within distributed Java EE environments using the Application Response Measurement (ARM) standard or similar means Aggregating data from multiple Java EE instances Modeling Java EE server clusters Supporting new component types: JSF, web-services, Message driven beans, Other Java EE 7.0 enhancements, e.g. Batch Jobs,.. 24 DevOps Performance WG

25 Thanks for your attention! Questions? Contact // Andreas Brunnert fortiss GmbH An-Institut Technische Universität München Guerickestraße München Germany tel: mail: fax: web: 25 DevOps Performance WG